顶级公司如何使用机器学习

在过去的一年里，机器学习热潮席卷了整个商业世界. According to Arthur Samuel, 半个世纪前创造了这个词的计算机科学家, machine learning is 定义 as the subfield of computer science that employs large data sets 和 training algorithms to “give computers the ability to learn without being explicitly programmed.”

Many executives have an intuitive sense that machine learning will prove as important a paradigm shift as the Internet 和 the personal computer. 最近的一次 调查 conducted by PwC indicated that 30% of business leaders believed AI would be the biggest disruption to their industry within five years. 仅在2016年，就有超过50亿美元的风险投资 淹没了 into machine learning startups. The McKinsey Global Institute 笔记 机器学习“广泛适用于许多常见的工作活动”,” including pattern recognition, 生成和理解自然语言, 和 process optimization.

为什么 Machine Learning Matters Now

最近的炒作是由三个关键发展推动的, which have reduced the barrier to entry for organizations across sector 和 stage that want to apply machine learning:

• More data 和 cheaper storage: The rise of cloud-based tools 和 the plummeting 成本 of storing data through services like Amazon Redshift mean that more data than ever is routinely generated 和 stored by business-critical applications.
• Open-source libraries: 广泛使用的机器学习库，比如谷歌的 TensorFlow 和 scikit-learn make cutting edge algorithms more accessible to a wider audience of data scientists 和 generalist software engineers.
• Greater horsepower: The development of cloud-based platforms 和 custom hardware optimized for machine learning means that these applications can run faster 和 at lower 成本, 增加它们对各种业务需求的适用性.

抽象地说，有令人信服的证据表明应该投资机器学习. 但是组织如何真正使用这项技术? 今天的机器学习是如何帮助企业创造价值的, cut 成本s 和 drive ROI?

In this article, we share case studies illustrating how companies of all sizes employ machine learning to address five key business cases: user acquisition, customer support, Forecasting, fraud prevention, 和 people management.

1. User Acquisition

In broad strokes, the customer acquisition funnel for a typical consumer or enterprise business has three stages: segmenting your customer base to underst和 和 address their needs, 在正确的时间用正确的信息吸引他们, 将他们转化为你产品的用户.

Machine learning has seen wide use by startups 和 major corporations alike across the entire user acquisition funnel. 亚马逊在2017年致股东的信中就是一个很好的例子 说 机器学习对亚马逊的贡献.com experience “beneath the surface” by powering product 和 deal recommendations based on user preferences. But segmenting users 和 showing them relevant products is only the first step: many retailers use machine learning to adjust br和ing, copy 和 promotional pricing on the fly to maximize the likelihood of a sale for any given customer.

在企业方面，Salesforce最近发布了 爱因斯坦, a product that examines CRM data to provide tailored recommendations to increase the chance that a particular prospect will convert from a sales pitch, 甚至建议发送电子邮件的合适时间.

2. Customer Support

当然，获取客户只是第一步. 无论是电子商务还是企业, 留住用户和限制流失需要提供及时有效的客户支持.

几十个品牌现在利用机器学习来改善客户支持体验. 例如，巴西超市Ocado 使用 Google machine learning APIs to 构建 a custom system that measures the sentiment of customer support inquiries 和 moves negative responses to the top of the support cue. 结果是Ocado对紧急信息的响应速度提高了四倍, 创造一个有价值的机会，在成为诋毁者的高风险中赢回客户.

最近，对话“机器人”现在是 筛选 support requests without help from a human operator—using machine-powered natural language to deliver a first response that can fulfill routine requests like issuing return labels. 除了 减少 support 成本s by up to 30%, 聊天机器人可以通过更快的响应来提高客户满意度, 随着他们理解能力的提高，他们的能力范围也会扩大. With a staggering 44% of U.S. 消费者 他更喜欢 与聊天机器人而不是人类互动, 投资于机器学习的面向消费者的企业将拥有巨大的优势.

3. Forecasting

In the back office, 各种各样的组织都开始使用机器学习来构建更健壮的系统, 精细而准确的Forecasting模型.

In 2016, Walmart ran a 竞争 数据科学招聘平台Kaggle, asking applicants to use historical data from 45 stores to 构建 a model that forecasted sales by department for each store. Insurance giant AIG assembled a 125 person data science team to 构建 machine learning models, 目标是提高公司Forecasting索赔和Forecasting结果的能力.

就连全球眼镜巨头陆逊梯卡也不例外 把 machine learning to work Forecasting dem和: it adds 2000 new styles to its collection every year, 并使用机器学习和过去发布的数据来Forecasting销售业绩.

4. Security 和 Fraud Detection

In 2016, fraud 成本 电子商务零售商平均超过总收入的7%. 对员工工资进行欺诈管理, 退款, 和 legitimate transactions that are denied due to false positives all contribute to this expense.

Machine learning is starting to bear out its potential as a powerful tool to intelligently monitor millions of transactions in real-time, 减少 waste from fraud. PayPal is a leader in this arena: they have 使用 open-source tools 和 their vast trove of transaction data to 构建 an artificial intelligence engine from scratch, 其主要目标是减少旧的欺诈模型产生的假警报的数量.

人类仍然在循环中训练模型并整理歧义, 但最初的结果是惊人的:自从实施了他们的新模式, 贝宝已经将其误报率降低了一半. 对于寻求白手套解决方案的公司来说，创业公司喜欢 筛选的科学 can consume a business’s data 和 apply fraud signals from their entire network of enterprise customers, 确保迅速抓住欺诈者的最新技术.

5. 人 Management

招聘、管理和留住高素质人才是所有业务能力的根本. One of the most onerous parts of hiring is filtering hundreds or thous和s of resumes to assemble a shortlist for interviews; over half of recruiters 说 这是他们工作中最困难的部分. 这个问题正在被像 Restless B和it, which makes a c和idate management system 使用 by companies like Adidas 和 Macy’s to filter resumes based on decisions that hiring managers have made in the past.

关键是，这些算法可以 训练有素的 to ignore unconscious human biases 和 even flag biased language in job descriptions—meaning that machine learning has the potential to identify high-performing, 在第一轮面试中可能被招聘人员忽视的多元化候选人. On the retention front, machine learning can augment the mentorship of great managers 和 help employees perform better by 生成 具体和公正的职业建议，基于过去的员工有类似的情况.

With a staggering 44% of U.S. 与人类相比，消费者更喜欢与聊天机器人互动, 投资于机器学习的面向消费者的企业将拥有巨大的优势.

机器学习的影响将越来越大

In this article, we have reviewed some of the most significant ways that machine learning can create direct 和 immediate value for a variety of organizations. 将机器学习视为某种企业的灵丹妙药是错误的, the performance of a machine learning system is only as good as the data on which it is 训练有素的, 和 an enterprise’s key decisions are often “edge cases” that require a measure of human judgement 和 anecdotal experience to assess.

而不是被机器学习的抽象潜力所迷惑, executives should approach the question of investing in this technology by taking stock of their core business challenges 和 matching them against the key capability of machine learning: drawing sense 和 meaning from a ton of data. 鉴于上述案例研究的多样性, 机器学习技术提供帮助的可能性可能比你想象的要大.